GitHub Copilot
The third-party benchmark aggregator benchgraph.dev lists Claude Haiku 4.5 (latest) explicitly with a SWE-bench Verified score of 73.3 dated 2026-04, placing it ahead of o3-pro (73.2) and Claude Opus 4 (72.7) but below Claude Mythos Preview (93.9), Claude Opus 4.6 (80.8), and Gemini 3.1 Pro Preview (80.6). The page is Beyond the SWE-bench entry, benchgraph.dev organizes benchmarks into categories (coding, knowledge, multimodal, agentic, math, reasoning, etc.) and reports that 112 models currently carry scores, with Claude Haiku 4.5 (latest) appearing among them as of April 2026. The page does not provide GitHub Copilot-specific avai